Research map / Vision-language and multimodal / CLIP and zero-shot
Training-Only Heterogeneous Image-Patch-Text Graph Supervision for Advancing Few-Shot Learning Adapters
Mohammed Rahman Sherif Khan Mohammad, Ardhendu Behera, Sandip Pradhan, Swagat Kumar, Amr Ahmed
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add focal-loss teacher supervision (vs CE only on final logits) (ablation) | TOGA · EuroSAT (16-shot) | accuracy: 88.1 -> 89.4 | Improved |
| use standard CE rather than Focal on teacher logits (ablation) | TOGA · EuroSAT (16-shot) | accuracy: 89.4 -> 87.5 | Hurt |
| add Top-N=50 node filtering instead of global pooling (ablation) | TOGA · EuroSAT (1-shot) | accuracy: 63.4 -> 67.4 | Improved |
| add Top-N node filtering instead of global pooling (ablation) | TOGA · Food101 (16-shot) | accuracy: 87.4 -> 87.5 | No clear effect |
| use multi-scale patches instead of fixed grids (ablation) | TOGA · EuroSAT (1-shot) | accuracy: 61.7 -> 67.4 (vs 3x3 grid) | Improved |
| add MGT in the teacher (ablation) | TOGA · EuroSAT (16-shot) | accuracy: 85.7 -> 89.4 | Improved |
| add MGT in the teacher (ablation) | TOGA · Food101 (16-shot) | accuracy: 87.2 -> 87.5 | No clear effect |
| add patch-text edges (ablation) | TOGA · EuroSAT (1-shot) | accuracy: 64.1 -> 67.4 | Improved |
| add unimodal Transformer enrichment (ablation) | TOGA · EuroSAT (1-shot) | accuracy: 66.2 -> 67.4 | Improved |
| replace Tip-Adapter-F-style training with TOGA graph-teacher supervision | Tip-Adapter-F · 11-dataset average (16-shot) | accuracy: 75.7 -> 82.3 | Improved |
| TOGA vs CCA on OOD ImageNet-A | TOGA · ImageNet-A | accuracy: 49.8 vs 50.9 (MaPLe) | Mixed |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- MaPLe — MaPLe: Multi-Modal Prompt Learning (CVPR 2023) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- SD-LoRA: Training-Time Structural Distillation into LoRA for Few-Shot Vision-Language Adaptation (NeurIPS 2026 · similarity 0.88)
- GraphAdapter: Tuning Vision-Language Models With Dual Knowledge Graph (NeurIPS 2023 · similarity 0.88)
- Beyond Graph Model: Reliable VLM Fine-Tuning via Random Graph Adapter (CVPR 2026 · similarity 0.87)
- UniGTE: Unified Graph–Text Encoding for Zero-Shot Generalization across Graph Tasks and Domains (NeurIPS 2025 · similarity 0.87)
- Mario: Multimodal Graph Reasoning with Large Language Models (CVPR 2026 · similarity 0.87)
- GraphVLM: Benchmarking Vision Language Models for Multimodal Graph Learning (CVPR 2026 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
